Scenarios
Healthcare scenario
Satisfaction ratings, symptom severity, wait times. Pre-anonymized + on-the-fly + insert flows.
GRRM · Laplace · GaussianFinancial scenario
Spending categories, brackets, transaction types. Also hosts the amount-mean real-vs-integer comparison.
GRRM · Laplace · GaussianTelemetry scenario
Feature usage, session duration, platform analytics.
GRRM · Laplace · GaussianSurvey scenario
Political leaning, income, education, age group.
GRRM · Laplace · GaussianMechanism deep dives
Correction estimator
Unbiased frequency recovery from LDP-perturbed histograms. Single-value estimate, full-distribution, and mean-from-frequencies.
GRRM (closed form) · Laplace (matrix inv.)Laplace mean: real vs integer
Same data, same ε, two units. Apply Laplace per row then average — see how the continuous mean and the bucket-index mean compare.
anon.ldp_laplace_real · anon.ldp_laplaceOne-hot vs scalar histogram
Two ways to estimate a histogram under Laplace/Gaussian LDP: scalar + correction, vs one-hot vector + noise + sum. Shows where OHE wins.
anon.ldp_{laplace,gaussian}_onehotExponential mechanism (central DP)
Pick the most common symptom-severity level under ε-DP. Run 1000 trials and plot how often the mechanism selects each candidate.
anon.ldp_exponential_mode (utility = count)DP query analysis (central DP)
Write an ordinary count(*) query; the extension derives sensitivity, adds Laplace noise, and refuses what it can’t bound. Declare a DP PID column to go user-level.
anon.dp_query · anon.dp_laplace